# 1. 克隆仓库
git clone https://github.com/TauricResearch/TradingAgents.git
cd TradingAgents

# 2. 创建虚拟环境
conda create -n tradingagents python=3.13
conda activate tradingagents

# 3. 安装依赖
pip install .

# 方式一：直接设置环境变量
export OPENAI_API_KEY=sk-xxxx      # OpenAI (GPT)
export ANTHROPIC_API_KEY=sk-ant-xxxx  # Anthropic (Claude)
export GOOGLE_API_KEY=xxxx         # Google (Gemini)
export XAI_API_KEY=xxxx           # xAI (Grok)

# 方式二：使用 .env 文件
cp .env.example .env
# 编辑 .env 填入你的 API 密钥

export ALPHA_VANTAGE_API_KEY=xxxx
# 免费注册: https://www.alphavantage.co/support/#api-key

tradingagents
# 或直接运行
python -m cli.main

from tradingagents.graph.trading_graph import TradingAgentsGraph
from tradingagents.default_config import DEFAULT_CONFIG

# 初始化交易图
ta = TradingAgentsGraph(
    debug=True,
    config=DEFAULT_CONFIG.copy()
)

# 执行分析
_, decision = ta.propagate("NVDA", "2026-01-15")
print(decision)

from tradingagents.graph.trading_graph import TradingAgentsGraph
from tradingagents.default_config import DEFAULT_CONFIG

# 自定义配置
config = DEFAULT_CONFIG.copy()
config["llm_provider"] = "openai"           # LLM 提供商
config["deep_think_llm"] = "gpt-5.2"        # 复杂推理模型
config["quick_think_llm"] = "gpt-5-mini"    # 快速任务模型
config["max_debate_rounds"] = 2              # 辩论轮数

# 初始化
ta = TradingAgentsGraph(debug=True, config=config)

# 执行分析
result, decision = ta.propagate("NVDA", "2026-01-15")

# 输出结果
print(f"交易决策: {decision}")
print(f"详细信息: {result}")

{
    "ticker": "NVDA",
    "date": "2026-01-15",
    "decision": {
        "action": "BUY",           # BUY / SELL / HOLD
        "position_size": 0.10,     # 10% 仓位
        "stop_loss": -0.08,        # -8% 止损
        "target": 0.15,            # +15% 目标
        "confidence": 0.78,        # 78% 置信度
        "reasoning": "..."         # 决策理由
    },
    "risk_assessment": {
        "risk_rating": 2,          # 1-5 风险评级
        "risk_factors": [...]
    },
    "analyst_reports": {
        "fundamentals": {...},
        "sentiment": {...},
        "news": {...},
        "technical": {...}
    },
    "debate_summary": {
        "bull_case": "...",
        "bear_case": "...",
        "consensus": "..."
    }
}

config = DEFAULT_CONFIG.copy()
config["llm_provider"] = "ollama"
# 具体模型名称请参考 default_config.py 和 Ollama 文档

ta = TradingAgentsGraph(debug=True, config=config)

# 安装 Ollama
# macOS: brew install ollama
# Linux: curl -fsSL https://ollama.ai/install.sh | sh

# 下载模型（如 llama3、mixtral 等）
ollama pull llama3:70b
ollama pull mixtral:8x7b

# tradingagents/graph/nodes/analysts/esg.py
from typing import Dict, Any
from tradingagents.graph.nodes.base import AnalystNode

class ESGAnalystNode(AnalystNode):
    """ESG 分析师：评估公司 ESG 表现"""

    name = "esg_analyst"

    def __init__(self, llm, debug: bool = False):
        system_prompt = """你是一名专业的 ESG（环境、社会、治理）分析师。
你的职责是评估公司在以下三个维度的表现：

环境（E）：
- 碳排放与气候变化应对
- 资源使用效率
- 污染与废物管理

社会（S）：
- 员工关系与劳工权益
- 产品安全与质量
- 社区关系与社会责任

治理（G）：
- 董事会结构与独立性
- 高管薪酬合理性
- 股东权益保护

请基于公开信息和数据分析，给出 1-5 分的 ESG 综合评分。"""

        super().__init__(system_prompt, llm, debug)

    async def analyze(self, ticker: str, date: str) -> Dict[str, Any]:
        prompt = f"""请分析 {ticker} 在 {date} 的 ESG 表现。
重点关注：
1. 最近的 ESG 相关新闻
2. 监管文件中的 ESG 披露
3. 同行 ESG 对比

输出格式：
- ESG 综合评分: X/5
- 环境评分: X/5
- 社会评分: X/5
- 治理评分: X/5
- 关键发现: [...]
- 风险提示: [...]"""

        response = await self.llm.ainvoke(prompt)
        return self._parse_response(response)

# tradingagents/graph/trading_graph.py

class TradingAgentsGraph:
    def __init__(self, debug: bool = False, config: dict = None):
        # ... 现有初始化代码 ...

        # 添加 ESG 分析师
        self.esg_analyst = ESGAnalystNode(
            llm=self.llm,
            debug=debug
        )

    def _build_graph(self):
        # ... 现有图构建代码 ...

        # 添加 ESG 分析边
        self.graph.add_edge(
            "analyst_team",
            "esg_analyst",
            lambda state: state.update({"esg_report": ...})
        )

from tradingagents.graph.nodes.researchers import BullishResearcher, BearishResearcher

class DynamicDebateResearcher:
    """动态辩论：直到达成共识或达到最大轮数才结束"""

    def __init__(self, llm, max_rounds: int = 5, consensus_threshold: float = 0.7):
        self.bullish = BullishResearcher(llm)
        self.bearish = BearishResearcher(llm)
        self.max_rounds = max_rounds
        self.consensus_threshold = consensus_threshold

    async def debate(self, analyst_reports: dict) -> dict:
        for round_num in range(self.max_rounds):
            # 多头提出论点
            bull_case = await self.bullish.argue(analyst_reports, round_num)

            # 空头提出论点
            bear_case = await self.bearish.argue(analyst_reports, round_num)

            # 检查是否达成共识
            consensus_score = self._calculate_consensus(bull_case, bear_case)

            if consensus_score >= self.consensus_threshold:
                return {
                    "consensus": True,
                    "final_case": self._merge_cases(bull_case, bear_case),
                    "rounds": round_num + 1
                }

        return {
            "consensus": False,
            "bull_case": bull_case,
            "bear_case": bear_case,
            "rounds": self.max_rounds
        }

# tradingagents/tools/yfinance_data.py
import yfinance as yf

class YFinanceDataSource:
    """Yahoo Finance 数据源（部分代码参考 TradingAgents 贡献者）"""

    def get_price_history(self, ticker: str, start: str, end: str) -> pd.DataFrame:
        stock = yf.Ticker(ticker)
        df = stock.history(start=start, end=end)
        return df

    def get_financials(self, ticker: str) -> dict:
        stock = yf.Ticker(ticker)
        return {
            "income_stmt": stock.income_stmt,
            "balance_sheet": stock.balance_sheet,
            "cashflow": stock.cashflow
        }

    def get_news(self, ticker: str, count: int = 10) -> list:
        stock = yf.Ticker(ticker)
        news = stock.news
        return news[:count] if news else []

from tradingagents.graph.nodes.risk_manager import RiskManagerNode

class AdvancedRiskManager(RiskManagerNode):
    """高级风控：加入更多风控维度"""

    async def assess_risk(
        self,
        trade_proposal: dict,
        market_conditions: dict,
        portfolio_state: dict
    ) -> dict:

        # 调用基础风控评估
        base_risk = await super().assess_risk(
            trade_proposal, market_conditions, portfolio_state
        )

        # 添加新的风控维度
        # 1. 仓位集中度检查
        concentration_risk = self._check_concentration(portfolio_state)
        base_risk["concentration_risk"] = concentration_risk

        # 2. 相关性风险
        correlation_risk = self._check_correlation(
            trade_proposal["ticker"],
            portfolio_state["positions"]
        )
        base_risk["correlation_risk"] = correlation_risk

        # 3. 尾部风险（黑天鹅事件）
        tail_risk = self._estimate_tail_risk(
            trade_proposal["ticker"],
            market_conditions
        )
        base_risk["tail_risk"] = tail_risk

        # 综合评分
        base_risk["total_risk_score"] = (
            base_risk["base_score"] * 0.4 +
            concentration_risk * 0.2 +
            correlation_risk * 0.2 +
            tail_risk * 0.2
        )

        return base_risk

# 运行所有测试
pytest

# 运行特定模块测试
pytest tests/test_analysts.py -v

# 运行风控模块测试
pytest tests/test_risk_manager.py -v

# 生成测试覆盖率报告
pytest --cov=tradingagents --cov-report=html

# 开发工作流
git checkout -b feature/your-feature-name
# ... 开发 ...
git commit -m "feat: 添加新功能描述"
git push origin feature/your-feature-name
# 在 GitHub 上创建 Pull Request